import json import re import docx import base64 from docx.document import Document from docx.text.paragraph import Paragraph from docx.parts.image import ImagePart from docx.table import _Cell, Table from docx.oxml.table import CT_Tbl from docx.oxml.text.paragraph import CT_P from collections import OrderedDict class DocxChapterExtractor(object): def __init__(self, docx_path): self.doc = docx.Document(docx_path) # 解析文档 def extract_chapter_info(self, text): """提取章节编号和标题""" pattern = r'^(\d+(?:\.\d+)*)\s+(.*?)(?:\s*\d+)?\s*$' match = re.match(pattern, text) chapter_num = None content = None if match: chapter_num = match.group(1) # '4.1' or '4' content = match.group(2).strip() # '外部接口需求' else: print(f"'{text}' no match") return chapter_num, content def if_valid_match(self, chaptera_name, text): pattern = r'^(\d+(?:\.\d+)*)\s+' + chaptera_name + r'(?:\s*\d+)?\s*$' return re.match(pattern, text) is not None def get_chapter_number(self, chapter_name): """获取目录结构""" directory = [] chapter_num = '' flag = False for paragraph in self.doc.paragraphs: if self.if_valid_match(chapter_name, paragraph.text) and 'toc' in paragraph.style.name: chapter_num, content = self.extract_chapter_info(paragraph.text) directory.append((chapter_num, content)) flag = True continue if flag and paragraph.text.startswith(chapter_num) and 'toc' in paragraph.style.name: num, content = self.extract_chapter_info(paragraph.text) directory.append((num, content)) return directory def build_hierarchy(self, chapter_body_list): """将线性章节列表转换为嵌套结构""" hierarchy = {} path = [] # 当前路径栈，存储章节号的字符串部分（如 ["4", "2"]） for item in chapter_body_list: # 处理不同格式的输入数据 if len(item) == 3: num, content, _ = item # 忽略第三个元素 elif len(item) == 2: num, content = item else: continue # 跳过无效数据 # 切割章节号为字符串列表（如 '4.2.1' -> ["4", "2", "1"]） parts = num.split('.') # 1. 回溯路径找到当前层级 while len(path) >= len(parts): path.pop() # 2. 逐级构建或定位父节点 current_level = hierarchy for i in range(len(path)): part = path[i] # 如果父节点不存在，自动创建占位节点 if part not in current_level: current_level[part] = { "number": ".".join(parts[:i + 1]), "title": "[未命名章节]", # 占位节点标题 "children": {} } current_level = current_level[part]["children"] # 3. 插入当前节点 current_part = parts[len(path)] # 当前层级的部分（如 "1"） if current_part not in current_level: current_level[current_part] = { "number": num, "title": content, "children": {} } # 4. 更新路径栈 path = parts.copy() return hierarchy def extract_title_ordinal(self, s): # 正则表达式匹配以括号结尾的字符串 pattern = r'^(.*?)\s*[(（](.*?)[)）]$' match = re.match(pattern, s) if match: # 提取标题并去除前后空格 title = match.group(1).strip() # 提取序号并去除前后空格 ordinal = match.group(2).strip() else: title = s ordinal = None return title, ordinal def build_json_tree(self, chapter_body_list): """直接生成树形JSON结构""" root = {"number": "", "title": "ROOT", "content": "", "children": []} node_map = OrderedDict() node_map[""] = root # 初始化根节点映射 for item in chapter_body_list: # 处理不同格式的输入数据 if len(item) == 3: num, chapter_name, chapter_content = item title, ordinal = self.extract_title_ordinal(chapter_name) elif len(item) == 2: num, chapter_name = item title, ordinal = self.extract_title_ordinal(chapter_name) chapter_content = "" else: continue # 跳过无效数据 parts = num.split('.') parent_node = root # 始终从根节点开始查找父级 for depth in range(len(parts)): current_num = ".".join(parts[:depth + 1]) if current_num not in node_map: new_node = { "number": current_num, "title": title if (depth == len(parts) - 1) else "[未命名章节]", "ordinal": ordinal if (depth == len(parts) - 1) else "", "content": chapter_content if (depth == len(parts) - 1) else "", "children": [] } parent_num = ".".join(parts[:depth]) parent_node = node_map[parent_num] parent_node["children"].append(new_node) node_map[current_num] = new_node parent_node = node_map[current_num] # 确保最终标题和内容正确 node_map[num]["title"] = title node_map[num]["ordinal"] = ordinal node_map[num]["content"] = chapter_content return root["children"][0] if root["children"] else {} def is_image(self, graph: Paragraph, doc: Document): """判断段落是否图片""" images = graph._element.xpath('.//pic:pic') # 获取所有图片 for image in images: for img_id in image.xpath('.//a:blip/@r:embed'): # 获取图片id part = doc.part.related_parts[img_id] # 根据图片id获取对应的图片 if isinstance(part, ImagePart): return True return False def get_ImagePart(self, graph: Paragraph, doc: Document): # 一行只能获取一个图片 """获取图片字节流，类型为bytes""" images = graph._element.xpath('.//pic:pic') # 获取所有图片 for image in images: for img_id in image.xpath('.//a:blip/@r:embed'): # 获取图片id part = doc.part.related_parts[img_id] # 根据图片id获取对应的图片 if isinstance(part, ImagePart): return part.blob return None def iter_block_items(self, parent, directory): def custom_serializer(obj): if isinstance(obj, bytes): return { '__type__': 'image', 'format': 'base64', 'data': base64.b64encode(obj).decode('utf-8') } return obj """ 根据目录匹配章节内容 parent: docx解析内容, 传入self.doc directory: 章节目录结构，例如[('4', '工程需求'), ('4.1', '外部接口需求'), ('4.2', '功能需求'), ('4.2.1', '知识库大模型检索问答功能')] """ if isinstance(parent, Document): parent_elm = parent.element.body elif isinstance(parent, _Cell): parent_elm = parent._tc else: raise ValueError("something's not right") i = 0 body_list = [] body = [] flag = False # 判断是否循环到章节标题 for child in parent_elm.iterchildren(): if isinstance(child, CT_P): paragraph = Paragraph(child, parent) if i < len(directory) - 1: if paragraph.text == directory[i][1] and 'Heading' in paragraph.style.name: flag = True continue if paragraph.text == directory[i + 1][1] and 'Heading' in paragraph.style.name: # body_list.append(body) new_tuple = directory[i] + ( json.dumps( body, default=custom_serializer, ensure_ascii=False, ), ) body_list.append(new_tuple) # print(new_tuple) body = [] i += 1 continue if flag: if self.is_image(paragraph, parent): body.append(self.get_ImagePart(paragraph, parent)) elif paragraph.text != '': body.append(paragraph.text) elif i == len(directory) - 1: if 'Heading' in paragraph.style.name: new_tuple = directory[i] + ( json.dumps( body, default=custom_serializer, ensure_ascii=False, ), ) body_list.append(new_tuple) break if self.is_image(paragraph, parent): body.append(self.get_ImagePart(paragraph, parent)) elif paragraph.text != '': body.append(paragraph.text) # print(body_list) # print(paragraph.text, '--------------->', paragraph.style.name) else: flag = False elif isinstance(child, CT_Tbl): if flag: table = [] for row in Table(child, parent).rows: # 获取每行的单元格文本 row_text = [cell.text for cell in row.cells] # 用制表符或其他分隔符连接单元格内容 table.append("\t".join(row_text)) body.append(table) return body_list def main(self, chapter_name): directory = self.get_chapter_number(chapter_name) # print(directory) chapter_body_list = self.iter_block_items(self.doc, directory) # print(chapter_body_list) # 构建层级结构 # hierarchy = self.build_hierarchy(chapter_body_list) # print(hierarchy) json_tree = self.build_json_tree(chapter_body_list) return json_tree if __name__ == '__main__': docx_path = 'test - 副本.docx' extractor = DocxChapterExtractor(docx_path) extractor.main('工程需求')